Our new work BeamFormer has been accepted at ACM MobiSys 2026.

BeamFormer is a transformer-based framework that generates a full beam spectrum from a few reference beam measurements, drastically reducing the overhead of beam management in 6G networks. It achieves a 6.7 dB gain in beam alignment over existing approaches and produces 1,600-beam spectra in just 1.8 ms.

Thanks to my advisor Prof. Kun Qian and co-authors Swastik Kanjilal and Ish Jain for their support and contributions.